Mike_ wrote on 2026-05-29, 15:48:

shevalier wrote on 2026-05-29, 13:28:

In PSU without an APFC, the main converter itself is the only high-frequency load. The APFC is essentially a step-up converter. […]
Show full quote

In PSU without an APFC, the main converter itself is the only high-frequency load.
The APFC is essentially a step-up converter.
Thus, this capacitor is sandwiched between two high-current sources of high-frequency interference, and it often ends up in a bad state, unlike in older topologies.
When it fails (due to capacitance and ESR), the APFC (MOSFETs, diode, and often the controller) goes burn, making the repair of such a PSU cost half as much as buying a new one.
The primary capacitors in this topology fail completely without any outward signs.

If you’ve taken apart a PSU with an APFC, you always need to carry out 2½ steps:
- Measure the voltage across the large capacitor. For a 400V capacitor, the reading should be 380V ± 2–3V (for other ratings, check the voltage dividers connected to the controller. However, it definitely cannot be less than 375 volts)
- Measurement of ESR
- Measurement of capacitance, but this will be calculated automatically from the previous step.
According to current trends, 1 watt ~ 1 μF.

P.S. Alas, this is now one of the most likely points of failure. 🙁
We need to check them and wait for the price of these components to drop tenfold.
https://eu.mouser.com/ProductDetail/Vishay-Ro … kERvpWojg%3D%3D

I don't have an ESR meter, so that's out of the question...

I have a Chieftec with a 390μF Teapo capacitor.
This capacitor has a capacitance of 360μF and... an ESR of 4Ω.
Although it should be less than 1Ω.
And, interestingly enough, this capacitor looks in perfect condition.

So I apologise, but this isn’t a question of appearance or brand. 🙁
Unfortunately, the ESR does need to be only measured.
There are indirect ways of measuring the ESR, for example using an oscilloscope calibrator.
But that’s such a faff.
Personally, I measure it using a sound card.
I’m also into audio, and I need the impedance curve itself, not just a single figure.

https://deforg-free-fr.translate.goog/Zmeter. … &_x_tr_pto=wapp

shevalier wrote on 2026-05-29, 16:30:

I have a Chieftec with a 390μF Teapo capacitor. This capacitor has a capacitance of 360μF and... an ESR of 4Ω. Although it shou […]
Show full quote

I have a Chieftec with a 390μF Teapo capacitor.
This capacitor has a capacitance of 360μF and... an ESR of 4Ω.
Although it should be less than 1Ω.
And, interestingly enough, this capacitor looks in perfect condition.

So I apologise, but this isn’t a question of appearance or brand. 🙁
Unfortunately, the ESR does need to be only measured.

Well I don't have the equipment to measure ESR so that's not an option no matter what. The options are either to leave current cap as it is, or just in case replace it with the unused chemi-con SMQ 330µF cap in the picture.

EDIT: Never mind, I do have a function generator and for these higher ESR caps it's actually feasible to use it and oscilloscope to measure ESR...

EDIT2: I guess the old cap looks a bit off, even though this hardly is a precision measurement... 😁

New should be fine I guess.

shevalier wrote on 2026-05-29, 13:28:

If you’ve taken apart a PSU with an APFC, you always need to carry out 2½ steps:
- Measure the voltage across the large capacitor. For a 400V capacitor, the reading should be 380V ± 2–3V (for other ratings, check the voltage dividers connected to the controller. However, it definitely cannot be less than 375 volts)

Btw, why is this? I'd imagine the exact value would depend on PSU design rather than be a fixed value...

shevalier wrote on 2026-05-29, 13:28:

In PSU without an APFC, the main converter itself is the only high-frequency load. The APFC is essentially a step-up converter. […]
Show full quote

In PSU without an APFC, the main converter itself is the only high-frequency load.
The APFC is essentially a step-up converter.
Thus, this capacitor is sandwiched between two high-current sources of high-frequency interference, and it often ends up in a bad state, unlike in older topologies.
When it fails (due to capacitance and ESR), the APFC (MOSFETs, diode, and often the controller) goes burn, making the repair of such a PSU cost half as much as buying a new one.
The primary capacitors in this topology fail completely without any outward signs.

If you’ve taken apart a PSU with an APFC, you always need to carry out 2½ steps:
- Measure the voltage across the large capacitor. For a 400V capacitor, the reading should be 380V ± 2–3V (for other ratings, check the voltage dividers connected to the controller. However, it definitely cannot be less than 375 volts)
- Measurement of ESR
- Measurement of capacitance, but this will be calculated automatically from the previous step.

Well said.
Though usually just measuring the capacitance and ESR should suffice in almost all cases. Primary caps with abnormally high ESR (less likely) or abnormally low capacitance (you'll be lucky to catch this as the cap starts to fail) should be changed immediately.

Now, I'm not sure how much this will help, but I have noticed that some of the better-built PSUs with APFC (usually industrial and/or older server stuff) typically include either an MPP (metalized PolyPropylene) cap or a high voltage ceramic cap... or both... installed in parallel with the PSU's primary electrolytic cap. I'm guessing the idea behind this is for these caps to take on some of the high frequency ripple current from the APFC and thus relieve stress from the primary electrolytic cap (which should allow it to run cooler internally, and therefore greatly extend its operational life.) Also, in cases when the primary electrolytic cap fails completely open-circuit (and this is a fairly common failure mode), the MPP or ceramic cap may save the primary-side silicon parts from blowing up by at least absorbing some of the energy from the APFC boost inductor.

As an experiment, I have started copying this by installing an MPP cap in all of my APFC PSUs. For the first few, I used 450V 2.2 uF MPPs... but now have switched to 0.1 uF 630V Panasonic MPPs that are actually meant for high frequency high ripple current circuits. I don't know how much this will help, but I suppose time will tell. At least so far, I have not seen any adverse effects by doing this to any of my APFC PSUs.

shevalier wrote on 2026-05-29, 13:28:

According to current trends, 1 watt ~ 1 μF.

1 uF per 1 Watt would be a dream-come-true... but I have not see this in any consumer PSU yet.
If you are lucky, you get maybe 3/4 uF per 1 Watt - and that's for the better built server and industrial PSUs meant to run 24/7.
For consumer stuff, I find 0.5 uF per 1 Watt to be the norm more or less. 500W PSUs with anything between 220 to 270 uF is very very common. Rarely I see 330 uF like in my "oldschool" Antec Earthwatts EA-500 (which I recapped and now equipped with a 450V 500 uF Nichicon because... I happen to have a few from scrap industrial boards. 😀 )

shevalier wrote on 2026-05-29, 13:28:

P.S. Alas, this is now one of the most likely points of failure. 🙁

Indeed.
Worst of all, the PSU manufacturer can do very very precise calculations how long these caps will last in "regular" everyday use. Thus, you can be almost assured that if a PSU manufacturer gives you warranty for X years, those primary caps are probably calculated to do X +1 to 3 years more... maybe!
Well, at least this seems to be the case with the newest PSUs.
For the older APFC PSUs, it seems manufacturers still weren't aware of that, so some PSUs were done quite well and probably will be fine for many years to come. But others, like the Corsair CX430, are known ticking clocks.

Mike_ wrote on 2026-05-29, 11:54:

Btw, I recapped my Enermax, which apparently is active PFC so would it be a good idea to replace the primary cap? I don't recognize its brand, either. I happen to have a couple of these chemi-con SMQ caps available if they would be suitable.

Well, it this specific case, you may actually be OK (or at least not any worse) with leaving that Hitachi cap alone as tehsiggi suggested. My personal observation is that Hitachi primary caps tend to do OK in APFC (even to this day... or so far at least.) And since your Enermax is only rated for 300 Watts output, that 180 uF rating is relatively adequate. If anything, it's the voltage rating I'd be more worried about, since some APFC circuits can sometimes have/produce really "nasty" voltage transients or spikes... which over time can damage the 400V -rated caps (which is why I prefer to see 420V or even 450V -rated caps). But I see your Chemicons are also rated for 400V, they may or may not necessarily do better. Their only redeeming quality is that they are physically bigger, which means they will run cooler internally and thus more likely to last longer... again, provided the APFC in this PSU isn't pushing some silly transient voltage spikes.

So weather to change them or not - I'll leave that up to you.
Generally, if these Chemicon SMQs are brand new and you have no way to check the ESR and capacitance of that old Hitachi HP3 cap and you really want to have a peace of mind that this PSU will be fine for the next 5-7+ years of use... then go ahead and replace it (but save that Hitachi cap in case you need it for other PSUs in the future, as it still might be OK or at least better than some no-name garbage cap brand.)

shevalier wrote on 2026-05-28, 08:18:

I can’t test it under a normal load, as thats only got 2 × 100 μF input capacitors left capacity(they used to be 120 μF back in the day, Hipro has achieved a significant reduction in costs).

They are not the only ones.
The Corsair CX500 I got a few weeks back has a single 400V 220 uF cap... that reads only 200 uF currently. At least with the 2x 100 uF caps in your HiPro, you get better combined ESR (or so I'd think!)
Either way, these capacities are nowhere near good enough for a proper hold-up time for a 500+ Watt PSU... but I suppose that is the new norm now with modern PSUs.

shevalier wrote on 2026-05-28, 08:18:

I’d like to fit 170–220 μF * 400 V, but I can’t find any with an 18 mm diameter.

Nichicon PT series might have a cap that fits, though I have not checked. All I know is this series was used in the Nichicon -made PS3 PSUs for the "fat" PS3's (3x 140 uF @ 420V rating.) Good source of caps, those Nichicon PS3 PSUs, since they have all kinds of problems and aren't worth fixing. While the capacity (140 uF) may not be what you are looking for, at you get a higher voltage -rated cap. For APFC, that's actually important. In my observation, 400V caps tend to not last as long as 420V or 450V caps.

shevalier wrote on 2026-05-28, 08:18:

“honest Chinese capacitors”

And they would name this brand: HA. HA. HA. 🤣

Mike_ wrote on 2026-05-22, 15:51:

I put two 2.2µF caps in parallel, I guess that should work too even if it doesn't look pretty. 😁

Heh, not sure why I didn't think to suggest that too. This is definitely the more optimal solution. Good work! 😀

shevalier wrote on 2026-05-22, 03:58:

If the power capacitors in the secondary circuits dry out, the power supply unit will usually simply stop working properly. If t […]
Show full quote

If the power capacitors in the secondary circuits dry out, the power supply unit will usually simply stop working properly.
If the standby capacitors dry out, the entire power supply unit usually burns out. Often, the motherboard goes with it.
And taking a power supply unit apart is hardly a pleasant experience.
I prefer to do it once and for all.

No, that depends largely on the 5VSB circuit's design.

For 5VSB circuits designed around a PWM IC (e.g. UC384x, TEA1507, and etc.) or PWM-FET combo IC (TinySwitch or TopSwitch family, ICE3A/b family, VIPer01/02, STR-A6259h, and etc.), failure of either the "startup" cap typically results in the circuit just getting stuck in a startup loop or "hick-up" mode and not produce any significant voltage on its output. Meanwhile, a failure of the output capacitors usually results in the output voltage just dropping low and going out of regulation with a load. Both of these are pretty safe failure modes and are very unlikely to damage the attached hardware.

Now for 5VSB circuits that use the "oldschool" 2-transistor self-oscillating design, these indeed have more nodes where they can fail catastrophically when their startup / "critical" cap fails. BUT!!! - Not all 2-transistor 5VSB designs have this "critical" startup cap. Furthermore, from the ones that actually do, the design may still be such that the circuit does not fail catastrophically.
I've actually reverse-drawn many 2-transistor 5VSB circuits (and still do when I take apart various old PSUs) just to see what differences there are. From this, all I can tell you is that the 2T 5VSB circuits that are most likely to fail catastrophically are those that have the "critical" startup cap *and* when they are also utilized in a PSU with forward-converter technology, where there is an IC on the primary side that needs to be powered by the 5VSB's primary auxiliary winding. Because of the latter, generally these 5VSB circuits have a larger "critical" startup cap in the range of 22 to 100 uF on the primary and also tend to be implemented with a MOSFET as the power switch device for the 5VSB. The PSUs that use this design are (note: this is not an exclusive list):
- Bestec ATX-250-12E <-- a PSU that is 100% known to fail catastrophically on the 5VSB and take all attached hardware powered by that line. These made a real bad name for HP PCs back in the P4/Athlon XP era.
- older Delta and FSP from late 90's and early 2000's <-- complete and very bad failures are rare with the Delta PSUs, but still possible. For FSP, generally a little more likely due to worse cap offerings.
- ThermalTake TR2-430W (XP-550-NP)... which is the same as the HEC / Orion HP585D <-- same in regards to failures as with the Delta PSUs above
- Deer/Allied/L&C/Solytech -based PSUs (e.g. many Premier) <-- failure of the 5VSB circuit is very common on these due to very crappy caps used, but a catastrophic failure that takes out the motherboard is rare
- Powmax LP-6100
- Hyper Type-R PSUs from mid-late 2000's
... and probably many more that escape my head now.

That said, even these PSUs that use a "large" startup cap on the primary for the 5VSB don't really need this kind of extra treatment, going "all in" with ceramics or other exotic solutions. If you really just want to avoid a catastrophic failure with these, put a known-good Japanese brand cap from a reliable series suitable for PSU use, and that should easily grant you 20+ years of extra life on the PSU's 5VSB. Me thinks that is enough already. But for the extra paranoid, you don't need a stack of ceramic caps - just a single 4.7 uF to 10 uF ceramic in parallel to the electrolytic cap there already would be enough to "hold things over" (i.e. 5VSB not blow up or go over-voltage), should the startup electrolytic cap fail.

shevalier wrote on 2026-05-22, 03:58:

I have never come across a single S370-based motherboard on which the +3.3 V power supply has been implemented properly.
- Either a regulated step-up DC/DC converter supplying 3.3 V to the entire board,
- or a separate regulator for the memory/chipset, with all slots powered directly from the PSU.

??
Most s370 boards I've ran into powered AGP and PCI slots directly from the PSU rails. RAM and bridges were typically linear regulator affair.
I've never seen a step-UP regulator on any board so far, only step-DOWN.

shevalier wrote on 2026-05-22, 03:58:

Usually, it’s some sort of odd design.
- Asus-style: DC/DC converters with a +5 V output; the PSU’s 3.3 V output isn’t used at all
- Gigabyte style: separate linear regulators for memory and expansion slots.

Don't think you can generalize ASUS with any style. I've seen them use PSU 5V rail really heavily on some boards and 3.3V rail on others... also, some with a lot of DC-DC almost exclusively (e.g. the P4SD and variants) while others were linear-reg galore.
Old Gigabyte are a bit more predictable and generally lin-reg for anything other than CPU... but even then, not always.

shevalier wrote on 2026-05-22, 03:58:

As a result, the power supply goes haywire from such an uneven load.

And it shouldn't - properly designed group-regulated ATX PSUs should not care too much about how their 3.3V line is being loaded. Only the 5V to 12V rail load should matter.
When I do my cross-load tests, I sometimes leave the 3.3V rail completely unloaded and sometimes I don't. In any case, it's never been a problem... well, except for that Cyberlink I mentioned that did what you did and produced a 3.3V rail directly from the main transformer without any further regulation circuit... and also a really crappy Raidmax that didn't have its linearly-regulated 3.3V rail properly compensated on the feedback loop, producing pretty terrible regulation.

shevalier wrote on 2026-05-24, 11:55:

From personal experience. […]
Show full quote

momaka wrote on 2026-05-19, 14:05:

Therefore, DO NOT use polymers.

From personal experience.

The FSP ATX-350PAF power supplies work perfectly with Capson LZ 1800 μF × 16 V × 13 mΩ capacitors (electrolytic capacitors with ultra-low ESR). These capacitors are not of particularly high quality in themselves, but their ESR is comparable to the figures in the datasheet, at least when it comes to new and genuine units.
There were no other 10mm diameter capacitors available. 🙁

This general-purpose polymer capacitor from Kemet has, in principle, very similar technical specifications. (10mOhm)
https://eu.mouser.com/datasheet/3/72/1/A750MW … 28M1CAAE010.pdf
I don't have capacitors of that value on hand, so I haven't specifically tested them.

The experiments were carried out on a Hipro HP-D5201AW PSU 520 W (APFC, 2 transistors forward , Champion CM6800TX PWM controller, group regulation).
I tried general use polymer caps 1500 µF 10V on the +5V rail and 2200 µF 6.3V caps on the +3.3V rail, and they work perfectly.
The 820 µF*16V capacitors work fine, but their capacitance is critically insufficient.
The power supply stability issues started precisely when I added five more capacitors to the two 820 µF capacitors to achieve the standard capacitance of 2200 µF*2.
Polymer capacitors have a very weak dependence of their ESR on capacitance. This means you can't replace 2200 µF (12 mOhm) with 3 x 680 µF, because 680 µF have an ESR of 13 or 14 mOhm, not 36 mOhm.

So, most likely, if you use the same number of polymer cap as electrolytic cap and the same capacity, there shouldn't be any problems.
But experimentation is necessary.

Indeed experimentation is necessary if you want to use polymers.
BUT!
If you re-read this thread a little further back, you should see that my comment in regards to avoiding polymer cap use was towards older PSUs, particularly those based on the half-bridge topology. Same goes for the "2x 2200 uF on the 3.3V and 5V rails as minimum" - actually, that one was specifically in reply to tehsiggi's Codegen PSU.
The common denominator here is that half-bridge topologies based around old voltage PWM ICs such as the KA7500 or TL494 or SG6109 tend to have a relatively low switching frequency - typically under 30 KHz. Therefore, they really DO NEED the larger capacity on the output. Very low or ultra-low ESR is not needed here because of the relatively slow circuit Fsw. As for whether these PSUs can take ultra-low ESR caps and/or polymers all depends on how the compensation for the feedback loop is implemented (more precisely, where the "poles and zeros" are located on the phase response plot, as these will dictate where the feedback loop is stable and where it isn't.) In some cases, the very or ultra-low ESR of the caps will produce its own "ringing" ripple/noise on the output... and if the feedback loop is not compensated to filter that out, it may amplify this noise back into the PWM IC and make the PSU further oscillate with that (unwanted) frequency... thus causing severe noise on the output rails and/or possibly make the PSU completely loose its regulation or not work at all. But if the compensation is more forgiving, then the polymers might just be "fine" and not cause any issues... though this *should* be checked with an oscilloscope. Again, the reason being is that the output toroid is an inductive element, and with the very low ESR of the output caps, it will tend to produce its own "ringing", seen to the attached hardware as unwanted noise on the output. I've actually run into this latter scenario with a CWT PSU that was an early variant of their "ISO" series. The ultra-low ESR caps I used caused extra noise on the linearly-regulated 3.3V rail, making some old motherboards unstable with this PSU. As soon as I switched out the caps for some GP (not even low ESR!) stuff, all of the issues went away on the same motherboards (and these are motherboards I have used with other recapped PSUs in my arsenal, so I know they are OK and not the cause of the issues themselves). I ended up ordering Nichicon VZ general purpose 105C -rated caps for some of the outputs. These are not really ideal for SMPS use... but in this case, they kind of were.

Anyways. The moral here is that if one wants to avoid cases like the above, then it really is best *NOT* to use polymers or ultra-low ESR caps if the original design did not use any (but again, the 5VSB circuit can generally be the exception here.)
This being said, one can sometimes infer which PSUs may use very/ultra low ESR caps (or polymers) and still be fine, like the case of your HiPro. Generally, this is done by looking at the topology of the PSU and also the type of core used on the output common inductor/toroid. If visible under the windings... light green core with one side painted blue is Micrometals type -52, which is generally used on PSUs with medium switching frequency... or at least higher than most half-bridge ones (though not always.) An all-black core is Micrometals type -45, which while more lossy than type -52, is regularly used with further higher Fsw PSUs due to lower cost and higher permeability (allowing a smaller core size per output power). PSUs that use these two cores (-52 and -45 type) are more likely to be fine with very low ESR caps like UCC KZE to KZH or Panasonic FR/FM/FS or Nichicon HV/HW or Rubycon ZL-... and -possibly- polymers too. Then there is one more toroid core commonly used - all blue. This one is pretty much used almost exclusively on the higher Fsw PSUs and much more likely to see some polymer caps on the output of the PSU already from the factory. And if the (PC) PSU doesn't have an output toroid, then it's probably a modern half-bridge LLC design, which in most cases already come with polymer capacitors on their output.

shevalier wrote on 2026-05-28, 08:18:

Let me tell you what. It seems that, with the help of Google’s Gemini AI, I’ve actually managed to set up the frequency compensation for the feedback loop.
I certainly wouldn’t have managed it on my own. 🙁

Yeah, we did a few very basic poles-zeros phase response calculations back when I was studying engineering... and I remember almost nothing from the process. All I remember is merely the concept of it. Hence why I choose not to mess with polymers on PSUs, unless absolutely necessary. Instead, I take the AI/LLM approach, but not the tech way with Google or some other actual AI/LLM. What I mean is I just follow the same methods as AI - observe/analyze a large set of PSUs and make my decisions about what should be OK and what shouldn't based on an "average" of those observations, much like an LLM... but just in my head's own analog thoughts. 😀

Mike_ wrote on 2026-05-29, 20:43:

shevalier wrote on 2026-05-29, 13:28:

If you’ve taken apart a PSU with an APFC, you always need to carry out 2½ steps:
- Measure the voltage across the large capacitor. For a 400V capacitor, the reading should be 380V ± 2–3V (for other ratings, check the voltage dividers connected to the controller. However, it definitely cannot be less than 375 volts)

Btw, why is this? I'd imagine the exact value would depend on PSU design rather than be a fixed value...

This is determined by the capacitor’s rated voltage.
This voltage must not be less than 365 V (rectified 230 VAC +10%); otherwise, the APFC will shut down at the peaks of the sine wave under unfavorable conditions.
If the voltage is lower, a significant portion of the stored energy is lost, as it is proportional to the square of the voltage. For 420–450 V capacitors, this voltage could be increased, but the technical data sheets provide standard calculations for 380 V, so most people do not bother with this. The chances of obtaining a voltage other than 380 V in a home use PSU are very slim.

In any case, the voltage must not exceed 400 V, as this will cause the capacitor to fail.
This is where the resistors in the voltage divider start to play a crucial role.
An extra couple of tens of kiloohms can already alter the voltage by 5–10 volts. This is due not only to the accuracy of the resistors, but also to the temperature coefficient of resistance (will reduce the voltage by 2–3 volts)and the voltage coefficient of resistance (will increase the voltage by 5–7 volts).
And the ripple caused by the converters will be 10 volts.
Generally speaking, power supply manufacturers would be happy to save money on capacitors by using them at "400–1" V, but this isn’t technically feasible. Consequently, the combination of all these factors results in a voltage of 380 V across the capacitor.

momaka wrote on 2026-05-29, 21:52:

As an experiment, I have started copying this by installing an MPP cap in all of my APFC PSUs. For the first few, I used 450V 2.2 uF MPPs... but now have switched to 0.1 uF 630V Panasonic MPPs that are actually meant for high frequency high ripple current circuits. I don't know how much this will help,

The bypass capacitor for the electrolytic capacitor should have a capacitance of at least 1% of the electrolytic capacitor’s capacitance (preferably 10%).
2.2 μF is an excellent choice; 0.1 μF is simply useless. Its impedance will be ten times greater than that of the electrolytic capacitor at a frequency of 100 kHz.
However, more often than not, there’s simply nowhere to fit them – there’s literally no space.

1 uF per 1 Watt would be a dream-come-true... but I have not see this in any consumer PSU yet.

https://www.techpowerup.com/review/?category= … ower%20Supplies
All power supply units supporting ATX 12 V Ver. 3.1

Nichicon PT series might have a cap that fits,

From what I can buy within a reasonably realistic timeframe (rather than 30–40 weeks)
NICHICON UPH 220 uF 400 В - $15 each *2 + $4 delivery.
On the one hand, they’re definitely genuine and I’ll receive them the day after tomorrow.
On the other hand, $35 is a third of the price of an OEM Seasonic (with a local ‘manufacturer’s’ label).

shevalier wrote on 2026-05-28, 08:18:

“honest Chinese capacitors”

And they would name this brand: HA. HA. HA. 🤣

🙁 But you’re absolutely right about that.

just a single 4.7 uF to 10 uF ceramic in parallel to the electrolytic cap there already would be enough

The voltage-capacitance coefficient for MLCC capacitors depends heavily on their size. With the same dielectric material, a stack of 10 1206 1μF capacitors will lose less capacitance than a single 10μF capacitor.
At 25 volts, it may retain as little as 2–3 μF.
And I removed these 100 psc 1 μF × 63 V capacitors from a single industrial circuit board. In other words, they were simply there and found a use.

??
Most s370 boards I've ran into powered AGP and PCI slots directly from the PSU rails.

That’s why I prefer the Aopen MX3S to the ASUS TUSL2C. This Aopen is simply a paragon of reliability and a lack of technical frills

Indeed experimentation is necessary if you want to use polymers.
BUT!
If you re-read this thread a little further back, you should see that my comment in regards to avoiding polymer cap use was towards older PSUs, particularly those based on the half-bridge topology.

Yeah, we did a few very basic poles-zeros phase response calculations back when I was studying engineering... and I remember almost nothing from the process. All I remember is merely the concept of it. Hence why I choose not to mess with polymers on PSUs, unless absolutely necessary. Instead, I take the AI/LLM approach, but not the tech way with Google or some other actual AI/LLM. What I mean is I just follow the same methods as AI - observe/analyze a large set of PSUs and make my decisions about what should be OK and what shouldn't based on an "average" of those observations, much like an LLM... but just in my head's own analog thoughts. 😀

It seems that the frequency compensation on this Hipro was set up rather oddly from the start, and copying what others have done isn’t much better.
Gemini AI identified a 4-pole compensation circuit in the PSH series CWT. Either a genius calculated it, or they simply optimised the model in the simulator.
But the is tedious.
- СM6800TX
- ‘get a lecture’
- TL431 acting as its controller
- get a lecture
- its frequency compensation
- getting closer. Add such-and-such an RC circuit.
And so, 40 minutes just to get close to the question that was of particular interest.
But he doesn’t lose the train of thought on the subject. And this is the free browser version. Not bad at all.

shevalier wrote on 2026-05-30, 06:25:

This is determined by the capacitor’s rated voltage. This voltage must not be less than 365 V (rectified 230 VAC +10%); otherwis […]
Show full quote

This is determined by the capacitor’s rated voltage.
This voltage must not be less than 365 V (rectified 230 VAC +10%); otherwise, the APFC will shut down at the peaks of the sine wave under unfavorable conditions.
If the voltage is lower, a significant portion of the stored energy is lost, as it is proportional to the square of the voltage. For 420–450 V capacitors, this voltage could be increased, but the technical data sheets provide standard calculations for 380 V, so most people do not bother with this. The chances of obtaining a voltage other than 380 V in a home use PSU are very slim.

In any case, the voltage must not exceed 400 V, as this will cause the capacitor to fail.
This is where the resistors in the voltage divider start to play a crucial role.
An extra couple of tens of kiloohms can already alter the voltage by 5–10 volts. This is due not only to the accuracy of the resistors, but also to the temperature coefficient of resistance (will reduce the voltage by 2–3 volts)and the voltage coefficient of resistance (will increase the voltage by 5–7 volts).
And the ripple caused by the converters will be 10 volts.
Generally speaking, power supply manufacturers would be happy to save money on capacitors by using them at "400–1" V, but this isn’t technically feasible. Consequently, the combination of all these factors results in a voltage of 380 V across the capacitor.

I get that it needs to be above what rectified 230V is and below where the capacitor blows up. And I guess it makes sense that nowadays most power supplies have very similar designs in PFC stage. However on this Enermax voltage across the primary cap is 372V *after* changing the cap. I suppose it's just from an era where there was more variance in PSU designs or something?

momaka wrote on 2026-05-29, 21:52:

Well said.
Though usually just measuring the capacitance and ESR should suffice in almost all cases. Primary caps with abnormally high ESR (less likely) or abnormally low capacitance (you'll be lucky to catch this as the cap starts to fail) should be changed immediately.

Actually my attempt to measure ESR indicated that the Hitachi cap's ESR might be above what it should be and capacitance below it. According to datasheets the Chemi-con cap should have both higher capacitance and ESR than the Hitachi. In my measurement it instead looked like Hitachi's ESR was higher than that of Chemi-Con's and that difference in capacitances was larger than expected. Of course, you probably shouldn't draw too much conclusions from this as it's a very inaccurate measurement, but still. 😀

momaka wrote on 2026-05-29, 21:52:

Well, it this specific case, you may actually be OK (or at least not any worse) with leaving that Hitachi cap alone as tehsiggi suggested. My personal observation is that Hitachi primary caps tend to do OK in APFC (even to this day... or so far at least.) And since your Enermax is only rated for 300 Watts output, that 180 uF rating is relatively adequate. If anything, it's the voltage rating I'd be more worried about, since some APFC circuits can sometimes have/produce really "nasty" voltage transients or spikes... which over time can damage the 400V -rated caps (which is why I prefer to see 420V or even 450V -rated caps). But I see your Chemicons are also rated for 400V, they may or may not necessarily do better. Their only redeeming quality is that they are physically bigger, which means they will run cooler internally and thus more likely to last longer... again, provided the APFC in this PSU isn't pushing some silly transient voltage spikes.

So weather to change them or not - I'll leave that up to you.
Generally, if these Chemicon SMQs are brand new and you have no way to check the ESR and capacitance of that old Hitachi HP3 cap and you really want to have a peace of mind that this PSU will be fine for the next 5-7+ years of use... then go ahead and replace it (but save that Hitachi cap in case you need it for other PSUs in the future, as it still might be OK or at least better than some no-name garbage cap brand.)

I changed the cap, and out of curiosity, I tried to measure ripple voltage across it. I'm not sure how well my oscilloscope's add function works at high voltages like this, but this is what I got...

Mike_ wrote on 2026-05-30, 10:45:

I get that it needs to be above what rectified 230V is and below where the capacitor blows up. And I guess it makes sense that nowadays most power supplies have very similar designs in PFC stage. However on this Enermax voltage across the primary cap is 372V *after* changing the cap. I suppose it's just from an era where there was more variance in PSU designs or something?

That’s a 5% difference in power .
Which chip is the power supply built on? It would be interesting to have a look at its datasheet.
Perhaps the resistors in the high-voltage divider have failed, or perhaps it was configured that way from the start.

Mike_ wrote on 2026-05-30, 10:45:

I changed the cap, and out of curiosity, I tried to measure ripple voltage across it. I'm not sure how well my oscilloscope's add function works at high voltages like this, but this is what I got...

The attachment differential.jpg is no longer available

If it’s under maximum load, then everything’s fine.
If it’s idling, I’d say that’s a bit too much.
And the shape is a bit odd, too.
Could you at least check the capacitance of the film capacitor (0.47–2.2 μF) located immediately after the diode bridge?

You shouldn’t have taken an oscilloscope measurement in the high-voltage section of the power supply – it’s dangerous both to your life and to the oscilloscope itself.
Well, unless you have a differential high-voltage probe.

I've always got modern PSU's for ATX motherboard acquisitions .
I come to the understanding that the watt label represents the "peak" power output and not always achievable because of power line conditions.
I always increase upon the "recommended" watt specification usually at about an extra 25% or nearest official offering.
Additional Surge and overload protection is a must.

zapbuzz wrote on 2026-05-30, 14:33:

I've always got modern PSU's for ATX motherboard acquisitions . I come to the understanding that the watt label represents the " […]
Show full quote

I've always got modern PSU's for ATX motherboard acquisitions .
I come to the understanding that the watt label represents the "peak" power output and not always achievable because of power line conditions.
I always increase upon the "recommended" watt specification usually at about an extra 25% or nearest official offering.
Additional Surge and overload protection is a must.

You’re absolutely right, and it’ll work if your PC components are compatible with the current ATX version of the power supply.
And then, for the price of a couple of cups of coffee, they’ll sell you an A7n (or whatever that nForce 2-based motherboard from Asus is called) with a top-of-the-range Barton processor and a Radeon 9800 XT graphics card.
It turns out that power supplies compatible with the ATX 1.0 standard are no longer manufactured.

shevalier wrote on 2026-05-30, 13:55:

That’s a 5% difference in power .
Which chip is the power supply built on? It would be interesting to have a look at its datasheet.
Perhaps the resistors in the high-voltage divider have failed, or perhaps it was configured that way from the start.

PFC controller seems to be UCC3817N. Other IC on the daughterboard is UC3842.

Datasheet says that typical output voltage is 385 to 420 volts, which starts to be a bit much for 400V cap, so perhaps they changed the resistor dividers a bit?

shevalier wrote on 2026-05-30, 13:55:

If it’s under maximum load, then everything’s fine. If it’s idling, I’d say that’s a bit too much. And the shape is a bit odd, t […]
Show full quote

If it’s under maximum load, then everything’s fine.
If it’s idling, I’d say that’s a bit too much.
And the shape is a bit odd, too.
Could you at least check the capacitance of the film capacitor (0.47–2.2 μF) located immediately after the diode bridge?

You shouldn’t have taken an oscilloscope measurement in the high-voltage section of the power supply – it’s dangerous both to your life and to the oscilloscope itself.
Well, unless you have a differential high-voltage probe.

It's fine, I didn't touch the entire thing when it was powered on. I attached the probes first - ground wires went to the metal chassis of the PSU, definitely not to capacitor itself - and only then I put the power cord in at wall and watched what was on oscilloscope's screen. The 1:10 probes I used are rated for +-500Vpp so they should be fine, and oscilloscope itself is also rated for +-400Vpp (and it should also see just one tenth of the voltage due to using 1:10 probes).

Also, it was not just idling, I had a Socket A board with Athlon XP as a load. But it indeed does look odd, I'd say it's just an artefact rather than a proper measurement.

The yellow cap in the picture is 0.33µF, but I can't read the red cap's value without desoldering it.

Mike_ wrote on 2026-05-30, 16:05:

PFC controller seems to be UCC3817N. Other IC on the daughterboard is UC3842.
Datasheet says that typical output voltage is 385 to 420 volts, which starts to be a bit much for 400V cap, so perhaps they changed the resistor dividers a bit?

Thick-film resistors can age in two ways: either through grain oxidation and an increase in resistance, or through grain sintering and a decrease in resistance. The resistors’ nominals might have simply drifted. 372V isn't actually that bad; it's best not to mess with those resistors.

The yellow cap in the picture is 0.33µF, but I can't read the red cap's value without desoldering it.

In film resistors with sputtered electrodes (not foil-type), areas of damaged coating form on these electrodes over time as a result of overloading, leading to a reduction in capacitance. Sometimes this reduction can be significant.
It is precisely this capacitor that is worth simply measuring for capacitance. It filters the ‘half-rectified’ mains supply, and the APFC controller attempts to track this voltage waveform.
Due to the loss of its capacitance, the voltage waveform across the main electrolytic capacitor may become distorted.
Or simply a distorted mains AC supply (for example, clipped peaks of the sine wave, like a trapezium).
Or design features of this power supply unit.

In short, if the this capacitor is in good state, just wipe off the dust in PSU, put it back together and get on with it. 😀

shevalier wrote on 2026-05-30, 17:06:

In film resistors with sputtered electrodes (not foil-type), areas of damaged coating form on these electrodes over time as a re […]
Show full quote

In film resistors with sputtered electrodes (not foil-type), areas of damaged coating form on these electrodes over time as a result of overloading, leading to a reduction in capacitance. Sometimes this reduction can be significant.
It is precisely this capacitor that is worth simply measuring for capacitance. It filters the ‘half-rectified’ mains supply, and the APFC controller attempts to track this voltage waveform.
Due to the loss of its capacitance, the voltage waveform across the main electrolytic capacitor may become distorted.
Or simply a distorted mains AC supply (for example, clipped peaks of the sine wave, like a trapezium).
Or design features of this power supply unit.

In short, if the this capacitor is in good state, just wipe off the dust in PSU, put it back together and get on with it. 😀

Well I still don't have a meter to measure its capacitance with. 😁

I'd say the waveform was an artefact of measurement, because it actually changed with Vdiv. This scope's add function probably just doesn't work accurately enough when input voltages to channels 1 and 2 are so large.

Mike_ wrote on 2026-05-30, 17:21:

shevalier wrote on 2026-05-30, 17:06:

In film resistors with sputtered electrodes (not foil-type), areas of damaged coating form on these electrodes over time as a re […]
Show full quote

In film resistors with sputtered electrodes (not foil-type), areas of damaged coating form on these electrodes over time as a result of overloading, leading to a reduction in capacitance. Sometimes this reduction can be significant.
It is precisely this capacitor that is worth simply measuring for capacitance. It filters the ‘half-rectified’ mains supply, and the APFC controller attempts to track this voltage waveform.
Due to the loss of its capacitance, the voltage waveform across the main electrolytic capacitor may become distorted.
Or simply a distorted mains AC supply (for example, clipped peaks of the sine wave, like a trapezium).
Or design features of this power supply unit.

In short, if the this capacitor is in good state, just wipe off the dust in PSU, put it back together and get on with it. 😀

Well I still don't have a meter to measure its capacitance with. 😁

I'd say the waveform was an artefact of measurement, because it actually changed with Vdiv. This scope's add function probably just doesn't work accurately enough when input voltages to channels 1 and 2 are so large.

Even some multimeters are capable of measuring capacitance. 😀

If you have a low-current (small-sized) capacitor of 0.33 μF (for example, rated at 50–63 volts), then the circuit consists of a generator, a resistor and a capacitor.
You observe this divider using an oscilloscope and measure how many volts remain across capacitor. You then connect the capacitor under test and see what has changed.

shevalier wrote on 2026-05-30, 17:32:

Even some multimeters are capable of measuring capacitance. 😀

If you have a low-current (small-sized) capacitor of 0.33 μF (for example, rated at 50–63 volts), then the circuit consists of a generator, a resistor and a capacitor.
You observe this divider using an oscilloscope and measure how many volts remain across capacitor. You then connect the capacitor under test and see what has changed.

Fine, I tested it by measuring RC time constant and then calculated capacitance. It had text F1.0K 400V, which I suppose means 1µF, and it measured very close to that. 😁

Mike_ wrote on 2026-05-30, 19:09:

shevalier wrote on 2026-05-30, 17:32:

Even some multimeters are capable of measuring capacitance. 😀

If you have a low-current (small-sized) capacitor of 0.33 μF (for example, rated at 50–63 volts), then the circuit consists of a generator, a resistor and a capacitor.
You observe this divider using an oscilloscope and measure how many volts remain across capacitor. You then connect the capacitor under test and see what has changed.

Fine, I tested it by measuring RC time constant and then calculated capacitance. It had text F1.0K 400V, which I suppose means 1µF, and it measured very close to that. 😁

So everything’s fine then.

By the way, if the main capacitor in your Enermax is 30 mm in diameter, you’re really in luck.
There are two ridiculous sizes – 25 mm snap-in terminals and 18 mm through-holes.
Because all the common capacitors are mainly 30 mm and 20 mm snap-in types. 🙁

shevalier wrote on 2026-05-31, 08:30:

By the way, if the main capacitor in your Enermax is 30 mm in diameter, you’re really in luck.
There are two ridiculous sizes – 25 mm snap-in terminals and 18 mm through-holes.
Because all the common capacitors are mainly 30 mm and 20 mm snap-in types. 🙁

Actually it *is* 25mm snap-in, so I guess I got lucky that I happened to have a suitable replacement on hand. 😀

On the other hand, looks like it wouldn't have been that expensive to get a replacement from LCSC if I didn't have a suitable part.

https://www.lcsc.com/product-detail/C1579430.html

Mike_ wrote on 2026-05-31, 10:19:

shevalier wrote on 2026-05-31, 08:30:

By the way, if the main capacitor in your Enermax is 30 mm in diameter, you’re really in luck.
There are two ridiculous sizes – 25 mm snap-in terminals and 18 mm through-holes.
Because all the common capacitors are mainly 30 mm and 20 mm snap-in types. 🙁

Actually it *is* 25mm snap-in, so I guess I got lucky that I happened to have a suitable replacement on hand. 😀

Personally, I spent a long time looking for a ‘capacitor-like’ replacement with exactly this diameter that I could get hold of within a reasonable timeframe.
On the PSH (-1) series by CWT (CFT-650-14CS from Chieftec), the distance between the heat sinks is exactly this. And no capacitor even a millimetre thicker will fit in there.

shevalier wrote on 2026-05-28, 08:18:

Man, why don't Chinese manufacturers just create their own brand of HCC - “honest Chinese capacitors”?
Something like, “Average quality, reasonable price—here's the datasheet.”
We’ve learnt how to make capacitors and promise they meet the datasheet specifications.

Taicon is really feeling neglected.

gdjacobs wrote on 2026-06-01, 04:08:

shevalier wrote on 2026-05-28, 08:18:

Man, why don't Chinese manufacturers just create their own brand of HCC - “honest Chinese capacitors”?
Something like, “Average quality, reasonable price—here's the datasheet.”
We’ve learnt how to make capacitors and promise they meet the datasheet specifications.

Taicon is really feeling neglected.

I mean, you don't even have to go Chinese to find reasonably priced but reliable good capacitors. There are Koeran brands like Samyoung, Samwha etc. that I don't mind using for any of my repair jobs..